Rationality in Belief, Desire, and Action: A Unified Approach
Peter Railton, April 2009
(1) Problem to be solved: Old transcendental puzzle: How is rational action
possible? First, we need a model of acting for a reason that does not lead to regress, or
posit a rational homunculus. This would be a model of potentially rational action, since it
would fit acting for bad as well as good reason. Given such a model, we could might
then be able to say how action of this kind might actually be rational, i.e., acting for good
reasons, or acting in a manner appropriately attuned to reasons.
(2) Worries about regress: Current models of the difference between mere
behavior and acting for a reason typically involve the agent “taking a consideration as a
reason” or “identifying herself with a reason”. But this leads to a regress, since such
“taking” or “identifying” is itself an action. And either this action is itself done for a
reason (so the puzzle re-arises), or it is not (so that at bottom there is only arbitrary
volition). So we have made no progress in solving the problem.
(3) Generic, minimal idea of rationality as attunement to reasons: Let’s start
with a fairly minimal idea. At the least, our rationality should help equip us to respond
aptly to reasons as such – to help identify potentially relevant reasons, to help distinguish
real from apparent reasons, to enable us to adjust our responses – beliefs, actions, etc. –
to these reasons, etc. We need to understand non-accidental, contentful guidance by
reasons, e.g., thanks to the proper functioning of a general capacity for rationality in
thought and action. An Aristotelian skill.
(4) Current supposed dualism: Rationalist models, which tend to lead to regress or
fiat (see (2)), and empiricist models, which tend to conflate acting for a reason with a
mere disposition to respond to a cue. Neo-Humeanism and “direction of fit”.
(5) Broad idea: Look to the philosophically under-explored third man in mental
faculty talk, affect. And notice that affect serves to orient both thought and action
toward features of the environment that call for attention. Moreover, such affective
responses can be more or less warranted or appropriate. E.g., fear. The root of rational
is ratio: proportional. Affect can be – or become through learning – more or less
proportional to one’s circumstances, e.g., in representing them accurately, evaluating the
possibilities they afford, and influencing action. This is congruent with the thesis of
affective primacy, at the foundation of contemporary “dual process” models of the
the basic point [is] that brains are always and automatically evaluating everything they perceive,
and that higher-order thinking is preceded, permeated, and influenced by affective reactions
(simple feelings of like or dislike) which push us gently (or not so gently) toward approach or
avoidance. [Haidt, 2007]
(6) The “this is just psychology” objection: Am I simply doing psychology, with no
normative relevance? But even “taking to be a reason” and “identifying with a reason”,
which have been seen as “sources of normativity” are themselves “just psychological
states”. So the challenge – for rationalist as well as empiricist – is to give an account of
how psychological states can have normative relevance. We need a workable,
psychologically realistic model or we are just whistling in the dark.
(7) Belief – the problem in the small: Belief is distinguished from other attitudes in
virtue of (i) its functional role, e.g., in action guidance, and (ii) its “aim of truth”. The
former looks “merely causal”, the latter looks irreducibly normative and higher-order.
(8) The functional picture:
Belief and desire … are correlative dispositional states of a potentially rational agent. To desire
that P is to be disposed to act in ways that would tend to bring it about that P in a world in which
one’s beliefs, whatever they are, were true. To believe that P is to be disposed to act in ways that
would tend to satisfy one’s desires, whatever they are, in a world in which P (together with one’s
other beliefs) were true. [Stalnaker, 1984]
(9) Are we at the mercy of our beliefs, whatever they are? No, but we cannot
require that a belief have no action-guiding authority unless we first judge it to have such
authority. Instead, beliefs, perceptions, etc. must be accorded default normative authority
in reasoning and action.
(10) Carnap: This is a feature, not a bug. Learning requires priors. Feedforward and
feedback. Rationality in belief is dynamic: Where does one go from here? David Lewis.
(11) Bayes/Jeffrey: Priors need not be “logical” and a priori. Default, defeasible
normative authority permits learning and unlearning. Attunement and Good’s theorem.
(12) Bootstrapping? My advice: Pull up, not down.
(13) Modeling belief: Belief is a compound, dynamic state, not a lump attitude.
Belief that R:
A degree of confidence (trust) in a representation R functions to regulate a degree of
expectation that things are or will be as R portrays them; and this degree of confidence in
turn is modulated by whether in subsequent experience this expectation is met or
(14) Some explanation: (i) Self-trust as foundation for belief: default, defeasible
priors + experience posteriors ; (ii) two sorts of naive “degree of belief”, distinct
from higher-order probability; (iii) the inter-modality of learning and unlearning – belief
primes learning; (iv) non-propositional objects; (v) inference and affective transfer (e.g.,
rules of inference); belief that vs. belief in (faith); (vi) “bottom up” learning and unlearning.
(15) Some irrationalities of belief – decoupling and dysregulation: (i) Phobias,
opaque expectation, and epistemic akrasia; (ii) wishful thinking and self-deception; (iii)
belief change without learning (depression, mania); (iv) spurious learning – affective
“bleed through”; (v) dogmatism, obsession, and failure of feedback unlearning. Churchill.
(16) “Taking as a reason” revisited: We needed a state that was “taking as a
reason”, that was more than mere opaque expectation. The compound, first-order
state (13) can play that role: it is an expectation “under [or through] an idea” we to
some degree. It is not a judgment, so there is no need for higher-order attitudes or
meta-representations to get this off the ground, though these can be built up from it.
(17) Belief “aims at truth”: Compound state (13) is also “evidence-sensitive” or
“intrinsically self-correcting”, without need for higher-order normative judgment. “Self-
tending” in response to new evidence. The macaque and “cognitive tuning”.
(18) Affective primacy and etymology: belief, truth. Emotional intelligence as the
attunement of affect.
(19) Simple evolutionary considerations and psychological realism.
(20) Desire – the problem in the small: Desire involves: (i) functional role; (ii)
“desirability characteristic”. The first seems merely causal, “animal”; the latter seems
irreducibly normative, higher-order. Frankfurt’s unwilling addict; Quinn’s “radio man”.
(21) The Metro and “La planète désire”.
(22) Desire as a representation-mediated wanting: erotic desire vs. sex drive;
gourmandisme vs. hunger. Kant:
The faculty of desire is the faculty to be, by means of one’s representations, the cause of
the objects of those representations. [MM 212]
(23) Modeling desire: desire is a compound, dynamic state, not a lump attitude:
Desire that R:
A degree of positive affect toward a representation R functions to regulate a degree of
positive motivation toward bringing about the state of affairs that R portrays; and this
degree of affect is subsequently modulated by whether actual experience of moving
toward or realizing R is better, worse, or in conformity with expectations arising from
the affective representation.
(24) Some explanation: (i) Default, defeasible attitude – feedfoward, feedback (recall
the macaque and its reward-based learning); (ii) two naïve senses of wanting – “I want to
eat (to return to health), even though I don’t want to eat (have no hunger)”; (iii) learning
and informed desire – incentive learning, “affective forecasting”, and miswanting – desire
primes learning; (iv) Aristotelian practical reason as “deliberative appetition” and
motivated cognition through affective transfer – means/end reasoning; (v) positive front
end (vs. the “itch” theory) – the allure of desiring; (vi) focused attention, “incentive
salience”, and normative force: the desirability feature; (vii) desire vs. wish.
(25) Some irrationalities of desire – decoupling and dysregulation: (i) opaque
urges and the radio man, compulsive desires; (ii) addiction and reluctant addiction –
liking vs. wanting, some psychological evidence; (iii) akrasia and “preference reversal”;
(iv) change without learning (depression and mania); (v) spurious desire and “affective
(26) “Taking as a reason”: We needed to find a state that could constitute “taking
as a reason” without involving an action or judgment, and yet was more than mere
opaque wanting or uncontrollable craving. This compound, first-order state can play
that role: it is wanting “under [or through] an idea” that attracts us – a “desirability
characteristic”. Higher-order normative attitudes not needed, though they can be built
up from it. Contrast the “stopping problem” for higher-order desire accounts.
(27) Desire “aims at the good”: Desire involves “desirability characteristic”, but
represented affectively (“I like the idea of taking in a movie”), not in terms of normative
judgments. Desire as “intrinsically self-correcting” or “aiming” without need for higher-
order cognition. Evolutionary picture: foraging for benefit, avoiding harm.
(28) Affective primacy and etymology: desire, good. Emotional intelligence as
attunement of affect.
(29) Simple evolutionary considerations and psychological realism. Note what
is always presupposed in talking of evolution and accuracy in perception and belief.
(30) Direction of fit revisited: Desire is often seen as an inappropriate object for
reason because it has only a “world-to-mind” direction of fit, and so no associated
notion of correctness, evidence, inference, or learning. Pure conation. Within the
compound state, there is also a “mind-to-world” direction of fit, supporting inference
(e.g., means/end reasoning) and an associated notion of correctness, evidence, and
(31) Action for a reason without regress, in real time: Putting the models
together to yield a notion of action for a reason as “self-organized” mental activity. No
homunculus. Intelligent guidance of behavior under a regulative idea that “presents” its
grounds on the surface. Anscombe’s question. Neurological models of deliberation,
choice, and volition. Recruitment of motivation in the service of a positively-valenced,
trusted representation. Something we could actually, regularly, fluently do. A skilled,
(32) Rational action as practical attunement: Aristotle:
… brutes have sensation, but no share in action. Pursuit and avoidance in the sphere of
appetition correspond exactly to affirmation and negation in the sphere of intellect … .
[C]hoice is deliberative appetition, it follows that if the choice is a good one, both the
reasoning must be true and the desire right; and the desire must pursue the same things
that the reasoning asserts. We are here speaking of intellect and truth in a practical
sense … the function of practical intellect is to arrive at the truth that corresponds to
right appetition. [NE 1113a20-28]
(33) Quasi-transcendental argument (decaffeinated Nagel): If we have something
like a belief-desire psychology, and if there are reasons to act of the kind we intuitively
sense there to be, then our psychology can equip us to respond to them.